sign in sign up
<<
Home , Diplexer, NTSC

Digital Audio for NTSC

Craig C. Todd Dolby Laboratories San Francisco

0. Abstract

A previously proposed method of adding a digital similarities between B-PAL and -NTSC indicated carrier to the NTSC broadcast channel was found to that the Scandinavian test results would apply in the be marginally compatible with adjacent channel U.S. Our original 1987 proposal was: operation. The technique also has some problems unique to broadcasters. Digital transmission A. QPSK carrier with alpha=0.7 filtering. techniques are reviewed, and a new set of digital transmission parameters are developed which are B. Carrier 4.85 MHz above thought to be optimum for digital sound with NTSC carrier. television. C. Carrier level -20 dB with respect to peak vision carrier level. I. Introduction Compatibility testing of that system has been It is very feasible to compatibly add digital audio to performed, and some television sets have been found the NTSC television signal as carried on cable on which the data carrier causes detectable television systems. Besides the marketing advantage interference to the upper adjacent video channel in a which can accompany the use of anything "digital", clean laboratory setting. It should be noted that with there are some real advantages to digital these problem sets, the FM aural carrier also caused , especially where the transmission path noticeable interference to the upper adjacent picture. is imperfect. Digital transmission is inherently The interference from data occurs into luminance, robust. While the coding of high quality audio into and manifests itself as additive between digital form theoretically entails a loss of quality, approximately 1 MHz and 1.4 MHz. The noise there is no further loss of quality when the digits are level in a problem TV is subjectively similar to the transmitted through an error-free channel. This is in noise level resulting from a video carrier-to-noise stark contrast to the transmission of analog audio, in ratio (CNR) of approximately 47 dB. While this which a "perfect" channel is required to avoid level of interference may not be detectable given the degradation. Unfortunately, the intercarrier sound current quality level of cable signals in consumers' channel used with broadcast television is somewhat homes, the potential of optical fibre to raise tlJ.e less than perfect, and inherently limits the sound quality level of cable TV warrants a reduction in the quality of the analog BTSC stereo system. level of interference by approximately 6 dB to be safe. Digital transmission techniques have matured to the point where they can be economically applied to The system should be compatible with broadcast 1 2 broadcast audio. Previous work • led to a proposal television as well as . We must for the addition of digital audio to the NTSC consider broadcast problems which can result from television broadcast signal. That proposal borrowed existing transmitter plant, and broadcast spectrum heavily from work performed in Sweden3 and allocations. These considerations also call for some Finland where a 512 kb/s QPSK carrier was changes to the previously proposed system. extensively tested with PAL system B. The

50- 1990 NCTA TECHNICAL PAPERS II. Digital Modulation Basics bits/sec/Hz. Fig. 2 shows a block diagram of this modulator which is simply a pair of double Figure 1 shows the spectrum of a random data AM modulators in quadrature. The double sideband stream with a data rate of 250k bits/sec. In a digital modulator can be considered to be a phase transmission the data is always intentionally modulator, generating phases of 0 and 180 degrees. scrambled so that the transmitted data appears The second modulator can be considered to generate random. Note the spectral nulls at multiples of the phases of 90 and -90 degrees. The combination of data rate (250 kHz, 500 kHz, etc.). From Nyquist the two phases will generate one of four phases, sampling theory we know that all information in this ±45 and ± 135 degrees. This form of modulation is signal may be gleaned from the first 125 kHz, i.e. typically called Quadrature Phase Shift Keying or QPSK.

0 Fig. 3 shows the constellation diagram for QPSK. There are four possible states of the RF carrier phase. We refer to a chosen state as a symbol. -10 Since there are four possible states per symbol we are transmitting two bits per symbol with a symbol ---c:l rate of 250k symbols/sec, for a total data rate of ~ -20 v ~ Q Axis :::=' Q.s -30 0 0 <

-40

-50 '-----'----L.-....l--.L..-----L---L-....1.----1 I Axis 0 125 250 375 500 625 750 875 1000 Frequency (kHz) Figure 1 Spectrum of 250 kb/s data

the Nyquist frequency which is IJ2 of the 250 kHz sample rate. If we brick wall filter this signal at the 0 0 Nyquist frequency of 125kHz and use it to modulate a carrier, we will have a double sideband modulated Figure 3 QPSK Constellation of 250 kHz, which will carry 250k bits/sec of data (1 bit/sec/Hz). If we add a second 500k bits/sec. If the low pass filters were not similar channel at the same frequency but in present, the phase would be instantaneously jumping quadrature, we will double the information in the between the four states shown, at a rate of 250 kHz. 250 kHz bandwidth RF signal to 500k bits/sec, or 2 In practice the actual signal will be continuously traversing between these points and the receiver will I do:to. sample the signal at the time when the signal passes through the constellation points. Modulo. ted Co.rrier The QPSK receiver is shown in Fig. 4. A carrier recovery loop coherently regenerates the transmitted carrier, which was suppressed in the modulator. Q do.to. Quadrature mixers demodulate the I and Q signals, Figure 2 QPSK Modulator low pass filters limit the effective receive bandwidth,

1990 NCfA TECHNICAL PAPERS- 51 1 • [ O

2 H(jw) _ oos { ;; [w- ~(~«)]}, ~(1-cx)!>w!>~(1 +ex)] (1) [ Ts Ts 0, [.,, ;}1 +«)]

excess bandwidth over a perfect brickwall (alpha==O) Eye filter. A filter meeting the amplitude response of Moni-tor Eq. 1, and having constant group delay (linear phase) will have no intersymbol interference.

I clo.to.

0.9 Modulo. tecl

Co.rrier 0.8

0.7 Q do.to. .,.. 0.6 :s ... 0.5 a"" Figure 4 QPSK Demodulator ~ 0.4

and comparators and clocked latches recover the 0.3 data. The latches are driven by a symbol timing 0.2 recovery circuit which reconstructs the transmit data clock at the proper phase so that the sampling time 0.1 is correct. 0.0 0 50 100 150 200 250 While QPSK can theoretically achieve a spectral Frequency (kHz) efficiency of 2 bits/sec/Hz, this is not achievable in Figure 5 Raised Cosine Nyquist Filters· practice since a brick wall filter is not realizable. The steepness of the filter determines the spectral efficiency. Steep filters ring, and as the spectral When we speak of the channel filter, we mean the efficiency of QPSK is raised by sharpening the filter function of the complete channel. This filters, the ringing of the individual data pulses includes the transmit lowpass filter, any increases. This ringing can cause the pulses to transmit RF filtering, receiver preselection and IF interfere with each other (intersymbol interference) filtering, and receive baseband lowpass filtering. It unless the ringing is controlled. If the filtering is also includes the effect of the transmission path done properly, each individual pulse waveform, which might have a non-flat response (or selective except for the pulse being detected, will pass through fade). Typically, all of the intended filtering is zero at the instant the receiver samples the designed into the system at IF or baseband, and all waveform. Nyquist filters provide the desired other circuitry is made wideband. The most precise characteristics. A commonly used class of Nyquist control over filtering is achieved with baseband filters are the "raised cosine" filters which have a lowpass filtering. It is much easier to make a frequency response as described in Eq. I , and shown precision 125 kHz lowpass filter than a precise 250 in Fig. 5. The alpha term determines the fractional kHz IF filter.

52- 1990 NCfA TECHNICAL PAPERS QPSK Spectral Efficiency = ~ bits I Hz (2) 1+a As alpha is reduced, spectral efficiency of QPSK is improved. Equation 2 shows the relationship between spectral efficiency and alpha. As alpha is reduced, the data waveform will have more ringing and the "eye" pattern will become more complex. The eye pattern is observed at the point where the analog waveform is sampled to recover the data. In the QPSK decoder block diagram (Fig. 4), the eye monitor point is shown for the I channel, and is just after the baseband lowpass filter. In Fig. 6 we see some examples of eye patterns for alpha=0.3, Figure 6c Eye Pattern, Alpha = 0.3 alpha=0.7, and alpha=l.O. The eye patterns become more complex as alpha is lowered and spectral efficiency is improved. More accuracy is required in implementation of low alpha systems. Any error in amplitude or phase linearity will rapidly degrade a complex eye. The eye will appear to close. Fig. 5d shows an alpha = 0.3 eye with a lot of closure due to poor filtering. Partial eye closure due to imperfect filtering leaves less margin against noise and interference.

+ • -- . - -·~·-r----··· ---. r- Figure 6a Eye Pattern, Alpha = 1.0

Figure 6d Alpha = 0.3, Poor Filter Accuracy

For optimal performance over a noisy channel, the amplitude portion of the filtering should be partitioned in equal portions between the transmitter I Figure 6b Eye Pattern, Alpha = 0. 7 and receiver i.e. the transmit filter and receive filters should both have an amplitude response equal to the square root of the nyquist filter response. The cascade of the two filter characteristics will then be a Nyquist filter. The combination of the filters should have linear phase (constant group delay). If

1990 NCfA TECHNICAL PAPERS- 53 both the transmit and receive filters individually have absolute eye opening (which would give the same linear phase, the combination will as well. performance in a noise However, it is also acceptable for the receive filter or interference environment), the peak level is 3 to have group delay variations and to compensate for times higher, or about 9.5 dB. For the same carrier these in the transmit filter. level, the 16 QAM signal is 9.5 dB less rugged than QPSK. Attempting to achieve good spectral efficiency with low alpha filtering requires the use of more accurate filters with good phase compensation. Filters may III. Partial Response Signaling be implemented with either analog or digital circuitry. Since we are starting with digital data, the Another attractive technique which can achieve use of a digital filter in the transmit side is attractive. improved spectral efficiency is known as "partial The receiver is more easily implemented with an response signaling". This method involves the use analog filter which does not require the use of A-D of a controlled amount of intersymbol interference converters with high sample rates and anti-aliasing that is removed by an extra processing step in the pre-filters. In a broadcast application we want to receiver. The simplest form of partial response keep the receiver cost low, so we want to use the involves performing a running average of adjacent minimal receive filter (without phase compensation). pairs of bits before driving the modulator. If we If the receive filter magnitude response accurately average a pair of bits which can each be + 1 or -1 matches the square root of a Nyquist response, then we get symbols with three levels: +2, 0 or -2. This the transmit filter can be implemented as an FIR 3-level signal still only contains 1 bit of information (finite impulse response) filter with an impulse per baseband symbol, but the spectral characteristics response which is simply the time reversal of the are changed. The spectrum of the data is changed impulse response of the receive filter. The cascade 0~.--.--,--,--,--,--.-~ of the filters will then have a symmetrical impulse response with constant group delay, and will have an -10 amplitude response equal to that of a Nyquist filter. The matched filter criterion is satisfied, so P'1 "' -20 performance in the presence of noise will be optimal. "'" All phase compensation is done in the transmit filter ~" so receiver cost is minimized. s"' -30 < In order to achieve spectral efficiencies of 2 -40 bits/sec/Hz or greater, QPSK must be abandoned and a higher level modulation method used. This can be -so~~~--~~--~~--~~ 0 125 250 375 500 625 750 875 1000 done by hitting the modulator filters (Fig. 2) with Frequency (kHz) multilevel data symbols. With QPSK, the I and Q Figure 7 Partial Response Spectrum data are simple binary symbols with two levels (1 and 0, or + 1 volt and -1 volt). If pairs of bits are fed to 2 bit D-A converters, we can form four level from that shown in Fig. 1 to that shown in Fig. 7. data (0, 1, 2, 3; or +3 volts, + 1 volts, -1 volts, -3 The first spectral null has moved from 250 kHz volts). QPSK extended to 4 level symbols is known down to 125 kHz. If we filter off all of the energy as 16 QAM (Quadrature ). 16 beyond 125kHz (which is relatively easy), we will QAM has a theoretical (with ideal brick wall achieve a spectral efficiency of 1 bit/sec/Hz at filtering) spectral efficiency of 4 bits/sec/Hz. baseband. The use of a pair of these partial response Unfortunately, the added spectral efficiency is system modulators in quadrature is known as QPRS. achieved at the expense of ruggedness. The eye The constellation diagram for QPRS is shown in pattern of 16 QAM is similar to that of three QPSK Figure 8. eyes stacked on top of one another. For the same

54- 1990 NCTA TECHNICAL PAPERS Q Axis

-ATTEN OdB • •

---4~-----+------

CE:NTER ee .. OOOMHz. SPAN l .. OOOMHz -RBW g.,OkHz *VBW 30Hz SWP :SO••o • • Figure 10 Comparative RF Spectra

Figure 8 QPRS Constellation alpha=O. 7 and alpha= 1.0. The transmit filters used The pairwise bit averaging is equivalent to a filter are FIR filters with a magnitude response equal to with an amplitude response equal to a cosine the square root of the raised cosine curve (for the function, as shown in Fig. 9. In order to have QPSK cases), or the square root of a cosine curve matched filtering, the bit averaging should not be (the QPRS case). The spectrum of QPRS is very done at the transmitter, but the cosine filter function attractive, but its eye pattern (Fig. 11) shows it to be should be divided equally between the transmit and less rugged than QPSK. Since there are two eyes, in receive filters. That is, both the transmit filter and order to achieve the same absolute eye opening (and the receive filter should have an amplitude response thus the same ruggedness), the peak power must be equal to the square root of the first quadrant of a increased by a factor of two, or 6 dB. For the same cosine. The transmit filter may be a digital FIR peak power, QPRS should be 6 dB less rugged. filter with an impulse response which is the time There are some mitigating factors though. reversal of that of the receive filter. •.. t ...•.• o.-~~~~~~~~~~~

-10

"'a -3o <

-40

-50L-L_~~~~~~~~~~ 0 25 50 75 100 125 150 175 200 225 250 . Frequency (kHz) Figure 11 PRS Eye Pattern Figure 9 PRS Cosine Filter

First, the PRS data stream spends half its time at the Fig. 10 shows the spectrum of the transmitted data 0 level (data = 1,0 or 0,1), and half at the +2 or- carrier for several modulation schemes. The 2 level (data = 0,0 or 1, 1). This means that half of narrowest spectrum is for QPRS, the second the time the carrier is modulated by zero, and half of narrowest is alpha=0.3 QPSK, and then we have the time it is modulated by the peak value. QPRS

1990 NCfA TECHNICAL PAPERS- 55 will have an RMS power level half that of QPSK for

the same peak level. For the same RMS power ATTEN 10d9 level, QPRS will thus lose only 3 dB of ruggedness compared to QPSK, and not the 6 dB that is lost with equal peak power levels. Second, since there is some redundancy in the PRS 3-level data stream, it is possible to detect and correct some errors without intentionally including redundant FEC data in the data multiplex. Coding gains of 1.5 to 2.0 dB can be achieved economically. The combination of these factors indicate that for comparable

carrier-to-noise ratios (CNR), QPRS is only about 1 CENTER 64.30MHx SPAN lO.OOMHz to 1.5 dB less rugged than QPSK against gaussian -Rew 3.0kHs -vaw 3DDH~ 5\lriP 3D••cr noise. QPRS is an attractive candidate if its narrow Figure 12 Ch3, Data, Ch4 Spectrum spectrum would solve our interference problem.

While QPRS requires some added complexity in the There is also an advantage in receiver complexity. receiver, the amount is within reason for a consumer If the first sound IF is widened slightly, the product. If the narrowness of the QPRS signal inter-carrier mixer used to recover the 4.5 MHz FM significantly reduces the potential for interference, sound will also recover the data sound carrier at 4.85 QPRS could be adopted for digital TV sound MHz. This point can be tapped and fed to the QPSK broadcasting without seriously impacting the receiver demodulator. If the data carrier is associated with cost. the upper channel, the inter-carrier offset will be -1.15 MHz, and additional IF filtering and another inter-carrier mixer will be required. Filtering will IV. Practical Broadcast Considerations have to be sufficient to reject the image at 1.15 MHz above vision. The choice of placing the carrier In the case of adjacent channel operation, the above the channel is a good one, and we originally proposed data carrier is placed above the FM sound agreed with it. carrier of one channel, and below the lower vestigial sideband of another channel. Figure 12 shows the Unfortunately, there are some serious problems with RF spectrum of TV Ch 3 and 4, with an Alpha = this choice for broadcasters in the U.S. The most 0.3 QPSK carrier inserted between the channels. In serious is that in the case of TV Ch. 6, the data the case of adjacent channel operation, it makes little carrier would lie at 88.1 MHZ, making both 88.1 difference whether we say that the data signal is and 88.3 unusable for FM broadcasting. The second assigned to the upper or the lower channel. The is that many transmitter installations are unable to BBC4 choose to allocate the signal to the lower easily broadcast the signal due to the notch channel. There are advantages to this choice. in use. Stations which use separate amplification for sound and vision would introduce the signal into the Depending on the actual vestigial sideband filtering sound transmitter. The sound and vision transmitters in the video modulator, there may be some overlap are combined in a device known as a "notch of video energy into the spectrum occupied by the " in which tuned cavities (creating notches) data signal. This will allow video to interfere with reflect the sound signal into the . There are the data, possibly causing data errors. In the case of no cavities tuned to the frequency of the data signal, broadcasting without adjacent channel operation, and the diplexer would need expensive modifications placing the data carrier at the top of the channel in order to add them. While this is not a technical means that ones own video will not interfere with problem, it is an economic one and could seriously ones own data. hamper the acceptance of the digital system by broadcasters.

56- 1990 NCfA TECHNICAL PAPERS Another problem, unique to cable, has to do with sync suppression scrambling. When sync information is carried on the FM sound carrier as pulse modulation, the spectrum of the FM sound carrier is widened and overlaps the data spectrum. The presence of the data may cause misbehavior in 0.6 the descrambler, and the presence of the sync information on the FM sound carrier may cause 0.4 errors in the demodulated data.

Choosing to place the data carrier below the vision carrier appears to be the correct choice for the U.S.

Broadcasters may add the signal to the vision Freq. re Video Carder transmitter just after the vestigial sideband filter, and Figure 13 Receiver VSB Nyquist Filters it will be amplified linearly along with the picture. At this frequency there will be no problem with the notch diplexer. With the carrier below the channel, the data spectrum down by 100 kHz reduces the there is no problem with a scrambled channel, interference into the upper video picture by although the same problem will exist if the FM approximately 6 dB on problem receivers. sound carrier on the lower adjacent channel is pulse modulated. VI. Interference of Data into FM Sound

V. Interference of Data into Picture Our initial studies of this system concentrated on the interference of data into the BTSC sound signal. We Placing the data signal on the lower sideband of our found that with QPSK and alpha=0.7 filtering, the own channel makes it even more critical to avoid data carrier could be placed 350 kHz above the interference into video because now we will interfere sound carrier. With alpha=0.3 filtering, the carrier with ourselves. The mechanism of interference is could be moved down. another 50 kHz so that it is imperfect VSB Nyquist filtering in the TV receiver. only 300 kHz above the FM sound carrier. The While it is easy to make a TV receiver reject the mechanism for interference into FM sound is data signal there are a wide variety of receivers in demodulation of the data signal as noise by the FM use, some of which have poor rejection. Fig. l3 detector. The frequency of the demodulated noise is shows the ideal receiver IF filter response. Also equal to the offset between the FM carrier and the shown is the kind of response that could be data signal. Since the data signal is wideband, the susceptible to interference. The problem is the 'tail' demodulated noise is also wideband. in the rolloff. If the filter goes right down to near zero response at 1 MHz away from the vision carrier Fig. 14 shows the demodulated composite BTSC FM the data signal will be fully rejected. If the filter has signal from DC to 200 kHz with no modulation on a tail, the data will not be rejected and may be any of the BTSC channels. The 15.7 kHz pilot is visible in the picture. clearly visible, along with L-R noise (produced by the BTSC compressor working at full gain) and the The only way to reduce the potential interference is SAP carrier at 78.7 kHz. Overlaid is a plot with the to: 1) reduce the data carrier level, and 2) move the data carrier turned on (using the parameters to be data carrier farther away from the video carrier (so specified in section VIII). The data carrier causes a as to get farther down on the slope of the receive slightly higher noise level starting at about 90 kHz, filter). The data carrier can be moved by either and at 140 kHz the noise level rises abruptly. A moving the center frequency or narrowing the data higher noise level at these should be of spectrum, or both. Moving the upper bandedge of no consequence in a properly designed mono, stereo,

1990 NCI'A TECHNICAL PAPERS- 57 *1'\TTEN lOdB VII. Interference of Picture into Data

The system adopted in Scandinavia, which our first proposal was based on, has problems with interference of upper adjacent vision into data. The lower edge of the vestigial sideband energy extends low enough in frequency to be received by the data demodulator tuned to the lower channel. Their solution was to modify the spectral mask of allowed ST/'\RT OH:z:: STOP 200.0k.H:::: energy in the lower sideband. Basically, this -RBW l.Ok.Hz -vew 30Hz SWP 20A..ac involves changing out all VSB filters in cable Figure 14 Spectrum from FM Demod, modulators for new ones with the cutoff moved up No Channels Modulated approximately 200 kHz in frequency. Our original proposal would have required this change in modulator filtering as well.

or SAP decoder, and of only minor significance to a When choosing new parameters to make the system Pro channel decoder. A poorly designed decoder more compatible with adjacent channel operation, it which inadvertently mixes components above 100 is very desirable to make the system operate wit."'l kHz down to baseband might suffer a slight existing modulator filters. The steps taken to lower degradation due to the addition of the data carrier, interference of data into picture will help to lower but this type of decoder would also suffer a interference of picture into data. premature noise degradation as CNR is reduced. Figure 15 shows the same plots, but with the Left and SAP channels fully modulated with a 400 Hz VIII. A More Compatible System sinewave. This figure confirms that the noise induced into the composite baseband signal is similar Our original proposal of alpha=0.7 QPSK placed whether the FM carrier is fully modulated or not. 350kHz above the audio carrier at a -20 dB level is The presence of non-linear FM modulation does not adequately compatible with BTSC audio, but could appear cause the noise to spread down to lower benefit from about 6 dB of improved compatibility audio frequencies. with video. This could be achieved by dropping the carrier frequency by 100 kHz, but that would

-ATTEN lOdB compromise compatibility with audio. Another way to achieve the 6 dB improvement is to narrow the carrier bandwidth as the frequency is dropped, keeping the lower data bandedge the same distance from the FM sound carrier. In theory, a 50 kHz drop in carrier frequency accompanied by a 100 kHz narrowing in bandwidth should meet our target. The narrowing could be done by staying with QPSK and changing the filtering from alpha=0.7 to alpha=0.3. The bandwidth will narrow by 0.4 x 250 = 100

STOP 2:00.0kH::c kHz, and the carrier frequency can be dropped 50 SWP 20uec kHz. Figure 15 FM Demodulated Spectrum, All Channels Modulated Unfortunately, when this is tried, the improvement in video interference is only about 3 dB. This is

58- 1990 NCfA TECHNICAL PAPERS because the more tightly filtered data has a higher present our margin against error is reduced, but we peak level, and the peaks are visible in the picture. can still operate error free. The higher peak level has cost us 3 dB of the expected improvement. Changing from QPSK to If the QPRS and QPSK carriers were used with the QPRS, using the same RMS signal level, does not identical carrier levels as defined above, the QPSK help, because QPRS inherently has a peak level eye wouid start out twice as open as the QPRS eye. which is 3 dB larger than the average level. Since we can operate QPRS with 3 dB more level (as defined above, or at the same level on an RMS In making these comparisons, we have found it basis), the QPRS eye in the receiver will be 70% (-3 useful to normalize our signal levels to the nominal dB) the size of the QPSK eye. This smaller eye will peak level, ignoring the overshoots. The eyes shown be more susceptible to interference from lower in Fig. 6 are normalized in this way, with a nominal sideband video information. Careful measurements peak level of ± 2 divisions. We measure the level show the -20 dB QPRS system suffers a 2 dB of the modulated carrier with a static data pattern interference penalty compared to the -23 dB QPSK which produces a pure carrier modulated with this system. QPRS looses out on both circuit costs and level. This is how we set the level in our original ruggedness. Despite the attractively narrow proposal. Using this method of measuring level, we spectrum of QPRS, QPSK is the superior solution. find that we can meet our target by lowering the Therefore, should stay with QPSK. carrier frequency by 50 kHz and using either alpha=0.3 QPSK with a level of -23 dB, or QPRS The new QPSK carrier frequency will be 1.2 MHz with a level of -20 dB. These two signals will have below the vision carrier frequency. This value is comparable RMS levels, and using non-redundant very close to 1/3 of the NTSC chroma error correction in the QPRS demodulator, similar frequency, which is 1.193182 MHz. This is a performance in the presence of gaussian noise (the convenient value to choose, because it can allow QPSK would actually be about 1.3 dB more rugged). future receiver circuits to use the chroma oscillator as a reference to demodulate the data, instead of a In order to choose which method to pursue, we have requiring a separate to be locked to to look at the relative costs, and the performance in the incoming carrier. the presence of interfering signals. QPRS will require a more expensive decoder. Since the eye is The data rate may also be locked to video. We need multilevel, two comparators instead of one are approximately 512 k bits/sec of data to use a low required to convert from the analog waveform back cost digital audio coding method based on adaptive to digital data. The level of the QPRS waveform at delta-modulation. It is desirable to use the same the comparator inputs is also critical, so a better clock frequencies that are used in satellite AGC is required. implementations of this audio system such as B-MAC and HDB-MAC. Those systems place The data is subject to interference from both the FM integral numbers of audio bits on horizontal lines and audio carrier, and lower sideband visual information. so have a direct relationship to video. It turns out The receive data filtering can adequately reject any that using 117 of the chroma frequency as a data interference from the FM carrier. Interference from clock gives us a total bit rate of 511.363 kHz. With video cannot be fully filtered out because it can be the simplest conceivable multiplex structure, the inband, depending on the VSB filter in the audio clock rate will be 13 times the horizontal scan modulator. In order to improve rejection of video rate, which is identical to clock rate used by the information, we have increased the order of the data BMAC systems. Using the chroma oscillator in the receiver baseband low pass filtering from a 3 pole to receiver as a data clock reference will save another a 5 pole filter. This change, along with the lowering crystal in the data timing recovery circuitry. of carrier frequency, allows our eyes to remain open even with the worst case interfering vision modulation present. With the worst case signal

1990 NCfA TECHNICAL PAPERS- 59 IX. The 1990 Proposal References

The following seems to be the optimum set of 1. Craig C. Todd, A Compatible Digital Audio parameters for an NTSC compatible digital carrier: Format for Broadcast and Cable Television, IEEE Trans. on Consumer Electronics, Vol. CE-33, No. A. QPSK carrier with alpha=0.3 filtering. 3, Aug. 1987, pp. 297-305.

B. Transmit filtering phase compensates for a 2. Craig C. Todd, Digital Sound and Data for specified receive filter. Broadcast Television - A Compatible System, NAB Proceedings, 41st NAB Engineering Conference, c. Carrier frequency 1.193182 MHz (locked to 1987. 1/3chroma) below video carrier. 3. Anders Nyberg, Digital Multi-Channel Sound for D. Carrier level -23 dB relative to peak vision Television, ICCE Digest, June 1987. carrier level. 4. A. J. Bower, Digital Two-Channel Sound for E. Data rate 511.363 kHz (locked to 117 , ICCE Digest, June 1987. chroma).

IX. Conclusion

Minor modifications have been made to the NTSC compatible system described three years ago. We explored a number of methods of improving compatibility and settled on a new set of specifications. The new parameters improve compatibility, reduce receiver cost, and simplify transmission of the signal for both broadcasters an cable operators.

60- 1990 NCfA TECHNICAL PAPERS